An innocent reader of the following x86 KVM code:

bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
{
        if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
                return true;
...

may get very confused: if APF mechanism is not enabled, why do we report
that we 'can inject async page present'? In reality, upon injection
kvm_arch_async_page_present() will check the same condition again and,
in case APF is disabled, will just drop the item. This is fine as the
guest which deliberately disabled APF doesn't expect to get any APF
notifications.

Rename kvm_arch_can_inject_async_page_present() to
kvm_arch_can_dequeue_async_page_present() to make it clear what we are
checking: if the item can be dequeued (meaning either injected or just
dropped).

On s390 kvm_arch_can_inject_async_page_present() always returns 'true' so
the rename doesn't matter much.
Signed-off-by: NVitaly Kuznetsov <vkuznets@redhat.com>
Message-Id: <20200525144125.143875-4-vkuznets@redhat.com>
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

Currently, APF mechanism relies on the #PF abuse where the token is being
passed through CR2. If we switch to using interrupts to deliver page-ready
notifications we need a different way to pass the data. Extent the existing
'struct kvm_vcpu_pv_apf_data' with token information for page-ready
notifications.

While on it, rename 'reason' to 'flags'. This doesn't change the semantics
as we only have reasons '1' and '2' and these can be treated as bit flags
but KVM_PV_REASON_PAGE_READY is going away with interrupt based delivery
making 'reason' name misleading.

The newly introduced apf_put_user_ready() temporary puts both flags and
token information, this will be changed to put token only when we switch
to interrupt based notifications.
Signed-off-by: NVitaly Kuznetsov <vkuznets@redhat.com>
Message-Id: <20200525144125.143875-3-vkuznets@redhat.com>
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

Revert "KVM: async_pf: Fix #DF due to inject "Page not Present" and "Page Ready" exceptions simultaneously"

Commit 9a6e7c39 (""KVM: async_pf: Fix #DF due to inject "Page not
Present" and "Page Ready" exceptions simultaneously") added a protection
against 'page ready' notification coming before 'page not present' is
delivered. This situation seems to be impossible since commit 2a266f23
("KVM MMU: check pending exception before injecting APF) which added
'vcpu->arch.exception.pending' check to kvm_can_do_async_pf.

On x86, kvm_arch_async_page_present() has only one call site:
kvm_check_async_pf_completion() loop and we only enter the loop when
kvm_arch_can_inject_async_page_present(vcpu) which when async pf msr
is enabled, translates into kvm_can_do_async_pf().

There is also one problem with the cancellation mechanism. We don't seem
to check that the 'page not present' notification we're canceling matches
the 'page ready' notification so in theory, we may erroneously drop two
valid events.

Revert the commit.
Reviewed-by: NGavin Shan <gshan@redhat.com>
Signed-off-by: NVitaly Kuznetsov <vkuznets@redhat.com>
Message-Id: <20200525144125.143875-2-vkuznets@redhat.com>
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

The current codebase makes use of the zero-length array language
extension to the C90 standard, but the preferred mechanism to declare
variable-length types such as these ones is a flexible array member[1][2],
introduced in C99:

struct foo {
        int stuff;
        struct boo array[];
};

By making use of the mechanism above, we will get a compiler warning
in case the flexible array does not occur last in the structure, which
will help us prevent some kind of undefined behavior bugs from being
inadvertently introduced[3] to the codebase from now on.

Also, notice that, dynamic memory allocations won't be affected by
this change:

"Flexible array members have incomplete type, and so the sizeof operator
may not be applied. As a quirk of the original implementation of
zero-length arrays, sizeof evaluates to zero."[1]

sizeof(flexible-array-member) triggers a warning because flexible array
members have incomplete type[1]. There are some instances of code in
which the sizeof operator is being incorrectly/erroneously applied to
zero-length arrays and the result is zero. Such instances may be hiding
some bugs. So, this work (flexible-array member conversions) will also
help to get completely rid of those sorts of issues.

This issue was found with the help of Coccinelle.

[1] https://gcc.gnu.org/onlinedocs/gcc/Zero-Length.html
[2] https://github.com/KSPP/linux/issues/21
[3] commit 76497732 ("cxgb3/l2t: Fix undefined behaviour")
Signed-off-by: NGustavo A. R. Silva <gustavoars@kernel.org>
Message-Id: <20200507185618.GA14831@embeddedor>
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

This reverts commit 5b494aea.
If unlocked==true then the vma pointer could be invalidated, so the 2nd
follow_pfn() is potentially racy: we do need to get out and redo
find_vma_intersection().
Signed-off-by: NPeter Xu <peterx@redhat.com>
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

Similar to VMX, the state that is captured through the currently available
IOCTLs is a mix of L1 and L2 state, dependent on whether the L2 guest was
running at the moment when the process was interrupted to save its state.

In particular, the SVM-specific state for nested virtualization includes
the L1 saved state (including the interrupt flag), the cached L2 controls,
and the GIF.
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

KVM_CAP_NESTED_STATE is now supported for AMD too but smm test acts like
it is still Intel only.
Signed-off-by: NVitaly Kuznetsov <vkuznets@redhat.com>
Message-Id: <20200529130407.57176-2-vkuznets@redhat.com>
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

The test is similar to the existing one for VMX, but simpler because we
don't have to test shadow VMCS or vmptrld/vmptrst/vmclear.
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

Many tests will want to check if the CPU is Intel or AMD in
guest code, add cpu_has_svm() and put it as static
inline to svm_util.h.
Signed-off-by: NVitaly Kuznetsov <vkuznets@redhat.com>
Message-Id: <20200529130407.57176-1-vkuznets@redhat.com>
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

This allows fetching the registers from the hsave area when setting
up the NPT shadow MMU, and is needed for KVM_SET_NESTED_STATE (which
runs long after the CR0, CR4 and EFER values in vcpu have been switched
to hold L2 guest state).
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

According to the AMD manual, the effect of turning off EFER.SVME while a
guest is running is undefined.  We make it leave guest mode immediately,
similar to the effect of clearing the VMX bit in MSR_IA32_FEAT_CTL.
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

The authoritative state does not come from the VMCB once in guest mode,
but KVM_SET_NESTED_STATE can still perform checks on L1's provided SVM
controls because we get them from userspace.

Therefore, split out a function to do them.
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

The L1 flags can be found in the save area of svm->nested.hsave, fish
it from there so that there is one fewer thing to migrate.
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

Now that the int_ctl field is stored in svm->nested.ctl.int_ctl, we can
use it instead of vcpu->arch.hflags to check whether L2 is running
in V_INTR_MASKING mode.
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

This bit was added to nested VMX right when nested_run_pending was
introduced, but it is not yet there in nSVM.  Since we can have pending
events that L0 injected directly into L2 on vmentry, we have to transfer
them into L1's queue.

For this to work, one important change is required: svm_complete_interrupts
(which clears the "injected" fields from the previous VMRUN, and updates them
from svm->vmcb's EXITINTINFO) must be placed before we inject the vmexit.
This is not too scary though; VMX even does it in vmx_vcpu_run.

While at it, the nested_vmexit_inject tracepoint is moved towards the
end of nested_svm_vmexit.  This ensures that the synthesized EXITINTINFO
is visible in the trace.
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

There is only one GIF flag for the whole processor, so make sure it is not clobbered
when switching to L2 (in which case we also have to include the V_GIF_ENABLE_MASK,
lest we confuse enable_gif/disable_gif/gif_set).  When going back, L1 could in
theory have entered L2 without issuing a CLGI so make sure the svm_set_gif is
done last, after svm->vmcb->control.int_ctl has been copied back from hsave.
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

Extract the code that is needed to implement CLGI and STGI,
so that we can run it from VMRUN and vmexit (and in the future,
KVM_SET_NESTED_STATE).  Skip the request for KVM_REQ_EVENT unless needed,
subsuming the evaluate_pending_interrupts optimization that is found
in enter_svm_guest_mode.
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

kvm_vcpu_apicv_active must be false when nested virtualization is enabled,
so there is no need to check it in clgi_interception.
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

The control state changes on every L2->L0 vmexit, and we will have to
serialize it in the nested state.  So keep it up to date in svm->nested.ctl
and just copy them back to the nested VMCB in nested_svm_vmexit.
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

Restore the INT_CTL value from the guest's VMCB once we've stopped using
it, so that virtual interrupts can be injected as requested by L1.
V_TPR is up-to-date however, and it can change if the guest writes to CR8,
so keep it.
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

In preparation for nested SVM save/restore, store all data that matters
from the VMCB control area into svm->nested.  It will then become part
of the nested SVM state that is saved by KVM_SET_NESTED_STATE and
restored by KVM_GET_NESTED_STATE, just like the cached vmcs12 for nVMX.
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

Allow placing the VMCB structs on the stack or in other structs without
wasting too much space.  Add BUILD_BUG_ON as a quick safeguard against typos.
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

This will come in handy when we put a struct vmcb_control_area in
svm->nested.
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

Use l1_tsc_offset to compute svm->vcpu.arch.tsc_offset and
svm->vmcb->control.tsc_offset, instead of relying on hsave.
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

Everything that is needed during nested state restore is now part of
nested_prepare_vmcb_control.
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

Split out filling svm->vmcb.save and svm->vmcb.control before VMRUN.
Only the latter will be useful when restoring nested SVM state.

This patch introduces no semantic change, so the MMU setup is still
done in nested_prepare_vmcb_save.  The next patch will clean up things.
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

When restoring SVM nested state, the control state cache in svm->nested
will have to be filled, but the save state will not have to be moved
into svm->vmcb.  Therefore, pull the code that handles the control area
into a separate function.
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

Unmapping the nested VMCB in enter_svm_guest_mode is a bit of a wart,
since the map argument is not used elsewhere in the function.  There are
just two callers, and those are also the place where kvm_vcpu_map is
called, so it is cleaner to unmap there.
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

vmx_load_mmu_pgd is delaying the write of GUEST_CR3 to prepare_vmcs02 as
an optimization, but this is only correct before the nested vmentry.
If userspace is modifying CR3 with KVM_SET_SREGS after the VM has
already been put in guest mode, the value of CR3 will not be updated.
Remove the optimization, which almost never triggers anyway.

Fixes: 04f11ef4 ("KVM: nVMX: Always write vmcs02.GUEST_CR3 during nested VM-Enter")
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

svm_load_mmu_pgd is delaying the write of GUEST_CR3 to prepare_vmcs02 as
an optimization, but this is only correct before the nested vmentry.
If userspace is modifying CR3 with KVM_SET_SREGS after the VM has
already been put in guest mode, the value of CR3 will not be updated.
Remove the optimization, which almost never triggers anyway.
This was was added in commit 689f3bf2 ("KVM: x86: unify callbacks
to load paging root", 2020-03-16) just to keep the two vendor-specific
modules closer, but we'll fix VMX too.

Fixes: 689f3bf2 ("KVM: x86: unify callbacks to load paging root")
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

The usual drill at this point, except there is no code to remove because this
case was not handled at all.
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

All events now inject vmexits before vmentry rather than after vmexit.  Therefore,
exit_required is not set anymore and we can remove it.
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

This allows exceptions injected by the emulator to be properly delivered
as vmexits.  The code also becomes simpler, because we can just let all
L0-intercepted exceptions go through the usual path.  In particular, our
emulation of the VMX #DB exit qualification is very much simplified,
because the vmexit injection path can use kvm_deliver_exception_payload
to update DR6.
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

In case an interrupt arrives after nested.check_events but before the
call to kvm_cpu_has_injectable_intr, we could end up enabling the interrupt
window even if the interrupt is actually going to be a vmexit.  This is
useless rather than harmful, but it really complicates reasoning about
SVM's handling of the VINTR intercept.  We'd like to never bother with
the VINTR intercept if V_INTR_MASKING=1 && INTERCEPT_INTR=1, because in
that case there is no interrupt window and we can just exit the nested
guest whenever we want.

This patch moves the opening of the interrupt window inside
inject_pending_event.  This consolidates the check for pending
interrupt/NMI/SMI in one place, and makes KVM's usage of immediate
exits more consistent, extending it beyond just nested virtualization.

There are two functional changes here.  They only affect corner cases,
but overall they simplify the inject_pending_event.

- re-injection of still-pending events will also use req_immediate_exit
instead of using interrupt-window intercepts.  This should have no impact
on performance on Intel since it simply replaces an interrupt-window
or NMI-window exit for a preemption-timer exit.  On AMD, which has no
equivalent of the preemption time, it may incur some overhead but an
actual effect on performance should only be visible in pathological cases.

- kvm_arch_interrupt_allowed and kvm_vcpu_has_events will return true
if an interrupt, NMI or SMI is blocked by nested_run_pending.  This
makes sense because entering the VM will allow it to make progress
and deliver the event.
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

Instead of calling kvm_event_needs_reinjection, track its
future return value in a variable.  This will be useful in
the next patch.
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

L2 guest hang is observed after 'exit_required' was dropped and nSVM
switched to check_nested_events() completely. The hang is a busy loop when
e.g. KVM is emulating an instruction (e.g. L2 is accessing MMIO space and
we drop to userspace). After nested_svm_vmexit() and when L1 is doing VMRUN
nested guest's RIP is not advanced so KVM goes into emulating the same
instruction which caused nested_svm_vmexit() and the loop continues.

nested_svm_vmexit() is not new, however, with check_nested_events() we're
now calling it later than before. In case by that time KVM has modified
register state we may pick stale values from VMCB when trying to save
nested guest state to nested VMCB.

nVMX code handles this case correctly: sync_vmcs02_to_vmcs12() called from
nested_vmx_vmexit() does e.g 'vmcs12->guest_rip = kvm_rip_read(vcpu)' and
this ensures KVM-made modifications are preserved. Do the same for nSVM.

Generally, nested_vmx_vmexit()/nested_svm_vmexit() need to pick up all
nested guest state modifications done by KVM after vmexit. It would be
great to find a way to express this in a way which would not require to
manually track these changes, e.g. nested_{vmcb,vmcs}_get_field().

Co-debugged-with: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: NVitaly Kuznetsov <vkuznets@redhat.com>
Message-Id: <20200527090102.220647-1-vkuznets@redhat.com>
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

Initialize vcpu->arch.tdp_level during vCPU creation to avoid consuming
garbage if userspace calls KVM_RUN without first calling KVM_SET_CPUID.

Fixes: e93fd3b3 ("KVM: x86/mmu: Capture TDP level when updating CPUID")
Reported-by: syzbot+904752567107eefb728c@syzkaller.appspotmail.com
Signed-off-by: NSean Christopherson <sean.j.christopherson@intel.com>
Message-Id: <20200527085400.23759-1-sean.j.christopherson@intel.com>
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

Restoring the ASID from the hsave area on VMEXIT is wrong, because its
value depends on the handling of TLB flushes.  Just skipping the field in
copy_vmcb_control_area will do.

Cc: stable@vger.kernel.org
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

Async page faults have to be trapped in the host (L1 in this case),
since the APF reason was passed from L0 to L1 and stored in the L1 APF
data page.  This was completely reversed: the page faults were passed
to the guest, a L2 hypervisor.

Cc: stable@vger.kernel.org
Reviewed-by: NSean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

pic_in_kernel(), ioapic_in_kernel() and irqchip_kernel() have the
same implementation.
Signed-off-by: NPeng Hao <richard.peng@oppo.com>
Message-Id: <HKAPR02MB4291D5926EA10B8BFE9EA0D3E0B70@HKAPR02MB4291.apcprd02.prod.outlook.com>
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>